Experts say exploiting Anthropic's Fable is not how Kimi K3 got so good

AI researchers pushed back on the White House's claim that Moonshot built Kimi K3 primarily by distilling Anthropic's Fable model, telling TechCrunch that the timeline and technical evidence do not support the theory. One expert told TechCrunch bluntly that they do not think you get a model this strong and this quickly on the heels of Fable doing strictly distillation, pointing out that Anthropic's Fable 5 only returned to public access on July 1, leaving Moonshot roughly two weeks before Kimi K3's July 16 launch, an unusually short window to build a genuinely frontier competitive model primarily by learning from another model's outputs. Distillation, the practice of training a smaller model using outputs generated by a stronger existing one, typically still requires substantial independent training infrastructure and time to be effective at the scale needed to produce a model as capable as Kimi K3 has proven to be across independent benchmarks, researchers say, making pure distillation an implausible sole explanation regardless of how much of Fable's output Moonshot may have had access to. The skepticism does not necessarily clear Moonshot of the White House's broader allegations, including the separate claim that the company accessed export controlled Nvidia chips through Thailand, but it does suggest that the specific distillation theory driving talk of sanctions may be resting on a weaker technical foundation than officials have publicly suggested, complicating the administration's case for treating Kimi K3's capabilities as primarily the product of misappropriated American technology rather than genuine independent progress.